The proposed research aims to provide effective, large-scale means for obtaining reliable information about drug-drug interactions (DDIs), by focusing on and utilizing the multiple distinct types of evidence used in reporting DDIs. DDIs are a significant cause of adverse drug reactions, leading to emergency room visits and hospitalizations. DDI research aims to link between molecular mechanisms that underlie interactions and their actual clinical consequences, through several types of evidence. We distinguish three types of DDI evidence that are often provided in the literature: in vitro, in viv, and clinical. In vitro studies investigate molecular mechanisms of interaction; In vivo studies evaluate whether these interactions impact drug exposure in human subjects; Clinical studies test whether drug interactions change the actual response to drugs (e.g. drug-efficacy or adverse drug reactions). As such studies span several disciplines, typically the three types of evidence are not simultaneously available or reported. Missing evidence along any of the three types, creates a knowledge gap that can hinder translational research. For instance, if adverse interaction effects are clinically observed, but the molecular underpinnings are not yet reported, it is difficult to identify a safe, alternative drug treatment. In this project we propose to develop and use large-scale text-mining methods and tools to mine drug- interaction information from PubMed abstracts and from FDA drug labels. These tools will be designed to explicitly identify gaps across the three levels of DDI evidence, and to help close such gaps. While automated discovery of DDI mentions in text is an active research area, no other text-based work is concerned with identifying explicit evidence for DDI, while separately taking into consideration the distinct types of interaction evidence. As a follow-up step, we also propose to conduct selective molecular pharmacology experiments to close the identified knowledge-gaps at the in vitro evidence level. Specifically: In Aim 1, we shall construct the needed lexica and new text corpora pertaining to in vitro, in vivo, and clinical DDI evidence; In Aim 2, a suite of text mining tools to separately identify the three types of DDI evidence will be developed, utilizing the corpora created in Aim 1; In Aim 3, clinically significant DDIs that have no sufficient in vitro evidence will be selected using the tools developed in Aim 2, and experiments will be conducted to evaluate in vitro metabolic enzyme- based DDI mechanisms. To the best of our knowledge we are the first group that sets out to distinguish among - and make use of - the different types of text-based DDI evidence in a systematic way. Following the text- based discovery with a selective molecular pharmacology experimental evaluation, is another unique interdisciplinary characteristic that adds to the significance of the proposed work. The successful completion of the proposed project will provide methods and tools for large-scale extraction of DDIs from the literature, along with their supporting evidence at the three distinct levels. Moreover, DDIs that will be reliably supported by one type of evidence but not another will be identified as strong candidates for future pharmacology research.